Crime Data Visualisation
Max Sharman
2024-03-22
Introduction
The quality of life for those living in Colchester is affected by Crime and Disorder [1]. The purpose of this report is to present data and uncover insights that may not be immediately apparent, and to offer data-driven guidance in reducing crime rates in Colchester in order to enhance the quality of life for its residents. The report makes use of advanced graphics and interactive plots to analyse the policing dataset ‘crime23.csv’, which is available from data.police.uk. The ‘crime23.csv’ dataset can be accessed by referring to [2].
The ‘crime23’ dataset, a comprehensive record of crimes in Colchester during 2023, forms the backbone of this report. It includes crucial details such as crime type, location, date in the Y-m format, and outcome status. The report begins with a broad analysis of this dataset, identifying intriguing patterns, before delving deeper into these for a more thorough examination.
A second dataset, temp2023.csv, from a weather station near Colchester, has been provided. The dataset contains daily climate data such as temperature, wind speed, pressure, and visibility. During the report, the climate data has been used to uncover relationships between the crime data and weather climate data. The ‘temp2023.csv’ dataset can be accessed by referring to [3].
# Reading Datasets into R studio
crime23 <- read.csv('crime23.csv')
temp2023 <- read.csv('temp2023.csv')Crime Frequency Analysis
To begin this report, a general analysis of crime frequency is performed using multiple data visualization techniques, such as tables, bar charts, a geographical map, a time series plot, and a seasonal boxplot. The purpose of this section is to comprehend the distribution of crime frequency data and identify areas of interest for further analysis.
The two-way table displays data on various types of crime that occurred in Colchester during 2023. This crime data includes shoplifting, anti-social behaviour, public order, and others. The data is presented in rows and columns to make it easier to understand. The table shows that there were 14 different categories of crime, with violent crime being the most frequent, having a frequency of 2633 cases. The second most frequent crime was anti-social behaviour, with 677 cases. The least frequent crime was possession of weapons, with only 74 cases.
attach(crime23)
# Two-Way-Table
ttable <- table(category, date)
two_way_table <- sort(table(category), decreasing = TRUE) # Sorting the data
# Kable to create a better view of the table
knitr::kable(two_way_table, caption = 'Colchester Crime Frequency in 2023', col.names = c('Crime', 'Frequency'))| Crime | Frequency |
|---|---|
| violent-crime | 2633 |
| anti-social-behaviour | 677 |
| criminal-damage-arson | 581 |
| shoplifting | 554 |
| public-order | 532 |
| other-theft | 491 |
| vehicle-crime | 406 |
| bicycle-theft | 235 |
| burglary | 225 |
| drugs | 208 |
| robbery | 94 |
| other-crime | 92 |
| theft-from-the-person | 76 |
| possession-of-weapons | 74 |
The bar chart presents the top 10 locations where crimes occurred most frequently. The x-axis represents the total number of offences committed, while the y-axis represents the categories of street names. The location ‘On or near’ had the highest number of total offences with 495 cases. However, this doesn’t provide the actual location of each crime. Therefore, data cleaning is required to fix this issue and assign each incomplete ‘On or near’ location to its actual location.
# Sorting data for the top 10 crime category occurrences
frequency_by_street <- crime23 %>%
group_by(street_name) %>%
summarise(total_offences = n()) %>%
arrange(desc(total_offences)) %>%
slice(1:10)
# Bar Chart
Offences_by_street <- ggplot(data = frequency_by_street, aes(x = street_name, y = total_offences)) +
geom_bar(stat = "identity", color = 'black', fill = "#5184C8") +
labs(title = "Top 10 Frequency of Offenses by Street Name",
x = "Street Name",
y = "Total Offenses") +
coord_flip() # Flipping bar chart for street name visibility
# Using ggplotly for interactive bar chart
ggplotly(Offences_by_street)This is a time series plot represented by a line graph, which shows the frequency of crimes committed per month. The blue line indicates the trend of crimes per month, showing whether the number of crimes goes up or down. The graph displays two notable peaks in January, with 651 reported crimes, and in September, with 642 crimes. However, there is a significant decrease from January to February, with only 467 reported cases in February.
# Creating a dataframe with the number of crimes per month
CrimeCounts <- table(crime23$date)
CrimeOffenceCounts <- as.data.frame(CrimeCounts)
names(CrimeOffenceCounts) <- c("Month", "CrimeCounts") # Renaming columns
# Creating time-series plot
Crime_per_month <- ggplot(CrimeOffenceCounts, aes(x = Month, y = CrimeCounts)) +
geom_point(color = "#5184C8", size = 3) + # Adding Points to graph
geom_line(aes(group = 1), color = "#5184C8", size = 1) + # Adding trend line
labs(title = "Number of Reported Offences per Month in 2023", # Titles
x = "Month",
y = "Number of Offences") ## Warning: Using `size` aesthetic for lines was deprecated in ggplot2 3.4.0.
## ℹ Please use `linewidth` instead.
## This warning is displayed once every 8 hours.
## Call `lifecycle::last_lifecycle_warnings()` to see where this warning was
## generated.# Using ggplotly for interactive time-series plot
ggplotly(Crime_per_month)This boxplot illustrates the number of crimes reported during each of the four seasons - Spring, Summer, Autumn, and Winter. Each season is represented by a different coloured box. The median line of Autumn and Winter appears to be at a similar level, indicating that crimes were most frequent during these seasons. Meanwhile, Summer and Spring have lower median bars, suggesting that the crime rate was comparatively lower during these seasons. Autumn, Spring, and Winter have similar quartile spreads, indicating higher variability, whereas spring has a smaller quartile spread.
# Creating new dataframes for each season by filtering the data by month
crime23_spring <- crime23 %>% filter(date %in% c('2023-03', '2023-04', '2023-05'))
crime23_summer <- crime23 %>% filter(date %in% c('2023-06', '2023-07', '2023-08'))
crime23_autumn <- crime23 %>% filter(date %in% c('2023-09', '2023-10', '2023-11'))
crime23_winter <- crime23 %>% filter(date %in% c('2023-12', '2023-01', '2023-02'))
# Combining dataframes with the added Season attribute
combine_seasons <- rbind(mutate(crime23_spring, Season = "Spring"),
mutate(crime23_summer, Season = "Summer"),
mutate(crime23_autumn, Season = "Autumn"),
mutate(crime23_winter, Season = "Winter"))
# Calculating the Crime frequency per season
season_crime_freq <- combine_seasons %>% group_by(Season, category) %>%
summarize(count = n()) ## `summarise()` has grouped output by 'Season'. You can override using the
## `.groups` argument.# Creating the boxplot
Seasonal_boxplot <- ggplot(season_crime_freq, aes(x = Season, y = count, fill = Season)) +
geom_boxplot()
labs(x = 'Seasons', y = 'Offence rate', title = 'Crime rates among different seasons') # Titles## <ggplot2::labels> List of 3
## $ x : chr "Seasons"
## $ y : chr "Offence rate"
## $ title: chr "Crime rates among different seasons"# Displaying boxplot
Seasonal_boxplot + scale_fill_manual(values = colour_palette) # Adding colour pallete
This map displays the locations of crimes in Autumn and indicates their frequency based on the density of each point. Darker and more dense points suggest that more crimes occurred in that area. The map shows that the centre of Colchester where more shops are located had more crimes as the circles in that area are darker and more dense. However, this map only provides a general overview of crime frequency and does not indicate the actual number of crimes committed in each area.
# Creating the map for crime locations in Autumn
map <- crime23_autumn %>%
group_by(street_name, category, lat, long)
m <- leaflet(map) %>%
addTiles() %>% setView(0.901681, 51.88924, zoom = 13.5) %>%
addCircleMarkers(popup = ~category, radius = 6, color = '#5184C8')## Assuming "long" and "lat" are longitude and latitude, respectivelymStreet/Area Name Analysis
In the previous section, I analysed the frequency graph of the street names in the ‘crime23.csv’ dataset and found that the data needed to be processed. This section will focus on resolving this problem and conducting a more detailed analysis of the ‘street_name’ attribute in the dataset.
Here, I handle the incomplete ‘On or near’ street_names. A table is presented displaying the street ids with missing locations and the number of missing street names they each have. This information is needed for matching the street id with its corresponding street/area name to fix the missing street/area names.
# Filtering data for 'On or near' street names
on_or_near <- crime23[crime23$street_name == "On or near ", ]
OnOrNearTable <- table(on_or_near$street_id) # Creating table
OnOrNearTable <- as.data.frame(OnOrNearTable) # Converting to a data frame
# Using Kable for a clear table
knitr::kable(OnOrNearTable, caption = 'Street Ids with Missing Locations', col.names = c('Street_Id', 'Freq'))| Street_Id | Freq |
|---|---|
| 2152922 | 15 |
| 2152938 | 3 |
| 2152957 | 6 |
| 2152961 | 5 |
| 2152969 | 41 |
| 2152976 | 3 |
| 2152989 | 3 |
| 2153007 | 6 |
| 2153018 | 23 |
| 2153025 | 24 |
| 2153040 | 1 |
| 2153051 | 17 |
| 2153068 | 4 |
| 2153081 | 1 |
| 2153105 | 23 |
| 2153111 | 26 |
| 2153123 | 77 |
| 2153143 | 4 |
| 2153155 | 37 |
| 2153173 | 35 |
| 2153190 | 10 |
| 2153232 | 24 |
| 2153245 | 2 |
| 2153280 | 1 |
| 2153297 | 3 |
| 2153318 | 9 |
| 2153373 | 24 |
| 2153395 | 1 |
| 2153436 | 27 |
| 2153520 | 26 |
| 2153615 | 3 |
| 2153667 | 11 |
I have displayed another table to identify the locations of the missing ‘On or near’ locations.
# Finding unique street IDs
unique_streets <- crime23 %>%
group_by(street_name) %>%
summarize(street_id = unique(street_id))## Warning: Returning more (or less) than 1 row per `summarise()` group was deprecated in
## dplyr 1.1.0.
## ℹ Please use `reframe()` instead.
## ℹ When switching from `summarise()` to `reframe()`, remember that `reframe()`
## always returns an ungrouped data frame and adjust accordingly.
## Call `lifecycle::last_lifecycle_warnings()` to see where this warning was
## generated.## `summarise()` has grouped output by 'street_name'. You can override using the
## `.groups` argument.# Filtering out street names containing only "On or near"
filtered_streets <- unique_streets[unique_streets$street_name != "On or near ", ]
# Displaying street IDs to locate missing locations
filtered_streets ## # A tibble: 374 × 2
## # Groups: street_name [350]
## street_name street_id
## <chr> <int>
## 1 Colchester Town (station) 2343256
## 2 On or near A133 2152980
## 3 On or near A134 2153054
## 4 On or near Abbey Field View 2153008
## 5 On or near Abbey Gate Street 2153113
## 6 On or near Albert Street 2153039
## 7 On or near Albion Grove 2153423
## 8 On or near Alexandra Road 2152967
## 9 On or near Alexandra Terrace 2152953
## 10 On or near Arnstones Close 2153790
## # ℹ 364 more rowsI created a ‘street_names’ vector to replace the incomplete ‘On or near’ location names, in order to identify their accurate locations. Then, I used the ifelse statement to update the street names based on the corresponding ‘street_names’ vector. Based on the tables that were created earlier, I was able to match all incomplete street names with their corresponding names by using their unique street IDs. However, there was one incomplete name that I could not identify, which I have labeled as “unknown” and left unresolved. The name is represented by the value “2153040”.
After that, I recreated the same bar plot from earlier that shows the fixed data. According to the bar plot, the most frequent crime location was ‘On or near Shopping Area’, with 480 crimes committed, followed by ‘On or near Supermarket’, with 376 crimes committed. From looking at the bar plot, I have successfully cleaned the data as the ‘On or near’ label has been removed. As crimes committed in shopping areas are the most frequent, I will be further investigating these areas.
# Creating a new vector to correctly adjust the street names
street_names <- c(
"2152922" = "On or near Shopping Area",
"2152938" = "On or near Parking Area",
"2152957" = "On or near Further/higher Educational Building",
"2152961" = "On or near Parking Area",
"2152969" = "On or near Police Station",
"2152976" = "On or near Parking Area",
"2152989" = "On or near Parking Area",
"2153007" = "On or near Nightclub",
"2153018" = "On or near Nightclub",
"2153025" = "On or near Shopping Area",
"2153040" = "Unknown",
"2153051" = "On or near Supermarket",
"2153068" = "On or near Parking Area",
"2153081" = "On or near Parking Area",
"2153105" = "On or near Conference/exhibition Centre",
"2153111" = "On or near Nightclub",
"2153123" = "On or near Shopping Area",
"2153143" = "On or near Petrol Station",
"2153155" = "On or near Parking Area",
"2153173" = "On or near Shopping Area",
"2153190" = "On or near Petrol Station",
"2153232" = "On or near Supermarket",
"2153245" = "On or near Bus/coach Station",
"2153280" = "On or near Supermarket",
"2153297" = "On or near Supermarket",
"2153318" = "On or near Parking Area",
"2153373" = "On or near Supermarket",
"2153395" = "On or near Shopping Area",
"2153436" = "On or near Supermarket",
"2153520" = "On or near Supermarket",
"2153615" = "On or near Sports/recreation Area",
"2153667" = "On or near Supermarket"
)
# Using an ifelse statement to adjust the street_name attribute
crime23 <- crime23 %>%
mutate(street_name = ifelse(street_id %in% names(street_names), street_names[as.character(street_id)], street_name))
# Filtering the fixed data to find the crime category frequency
frequency_by_street <- crime23 %>%
group_by(street_name) %>%
summarise(total_offences = n()) %>%
arrange(desc(total_offences)) %>%
slice(1:10) # Sorting for top 10
# Creating the new bar chart
Offences_by_street <- ggplot(data = frequency_by_street, aes(x = street_name, y = total_offences)) +
geom_bar(stat = "identity", color = 'black', fill = "#5184C8") +
labs(title = "Top 10 Frequency of Offences by Street Name",
x = "Street Name",
y = "Total Offenses") +
coord_flip() # Flipping bar chart for readable street names
# Using ggplotly to generate an interactive graph bar chart
ggplotly(Offences_by_street)Shopping Area Analysis
I have conducted additional analysis on street/area names and found that crimes occur most frequently in shopping areas. As a result, I have decided to further investigate these shopping areas. I have generated a bar plot that displays the most frequently occurring crimes in shopping areas. From the plot, it is evident that shoplifting is the most common crime with 211 offences, followed by violent crimes with 78 occurrences.
# Filtering the data for crimes located on or near shopping areas
ShoppingArea <- crime23 %>% filter(street_name %in% c('On or near Shopping Area'))
# Grouping by crime category and counting the frequency for each one
ShoppingAreaCount <- ShoppingArea %>%
group_by(category) %>%
summarize(count = n()) %>%
arrange(desc(count)) # Sorting in decending order
# Generating the bar plot
Offences_by_ShoppingArea <- ggplot(ShoppingAreaCount, aes(x = category, y =count)) +
geom_bar(stat = "identity", color = 'black', fill = "#AAF1E6") +
labs(title = "Frequency of Offences by ShoppingArea",
x = "Street Name",
y = "Total Offenses") +
coord_flip() # Flipping the graph so that the crime categories are readable
# Using ggplotly to generate an interactive graph bar chart
ggplotly(Offences_by_ShoppingArea)This pie chart shows the percentage distribution of crimes that occur in shopping areas. Each category of crime is represented by a different colour, and the size of each slice in the chart corresponds to the percentage distribution of the specific category. The largest slice, coloured in cyan, represents shoplifting, which accounts for 44% of all crimes committed in shopping areas. This is significantly higher than the other categories. The second-largest slice, which accounts for 16.3% of the total, represents violent crime. This is an interesting finding because violent crime was found to be the most frequent crime overall, but it is not the most common type of crime in shopping areas. Plotly documentation used for code can be accessed by referring to [4].
# Grouping Shopping areas by crime category and counting their occurrences
ShoppingAreaPie <- ShoppingArea %>%
group_by(category) %>%
summarize(count = n())
# Calculating the sum of all crimes
TotalCount <- sum(ShoppingAreaPie$count)
# Calculating the percentage distribution for each crime
ShoppingAreaPie <- ShoppingAreaPie %>% mutate(Percentage = round((count / TotalCount) * 100, digits = 2))
# Using plot_ly to create the interactive pie chart
plot_ly(ShoppingAreaPie, labels = ~category, values = ~Percentage, type = 'pie', marker = list(colors = colour_palette)) %>%
layout(title = "Percentage Distribution of Crime In Shopping Areas",
showlegend = TRUE)This time series plot displays the monthly shoplifting offences within shopping areas. The red line represents the trend of the data, while the interactive data points indicate the total number of shoplifting offences for each month. There was a significant decline in shoplifting cases from January to February. After that, there was a gradual decrease in the number of shoplifting offences from March to July. However, there was a significant increase from July to August, with a peak of 26 total offences in August. From August to December, there was another gradual decrease, and the lowest number of shoplifting offences (10) was recorded in December.
# Filtering ShoppingArea to only include shoplifting crimes
ShoppingAreaShoplifting <- ShoppingArea %>% filter(category == 'shoplifting')
# Creating a dataframe containing the respective crime occurrences for each month
ShoppingAreaShopliftingCounts <- table(ShoppingArea$date)
ShoppingAreaShopliftingCounts <- as.data.frame(ShoppingAreaShopliftingCounts)
names(ShoppingAreaShopliftingCounts) <- c("Month", "ShopliftingCounts") # Naming columns
# Generating the line plot for Shopping Area shoplifting counts
ShoppingAreaShopliftingMonthly <- ggplot(ShoppingAreaShopliftingCounts, aes(x = Month, y = ShopliftingCounts)) +
geom_point(color = "#F53646", size = 3) +
geom_line(aes(group = 1), color = "#F53646", size = 1) +
labs(title = "Number of Reported Shoplifting Offences per Month in Shopping Areas during 2023",
x = "Month",
y = "Number of Offences") +
theme_minimal()
# Using ggplotly to display an interactive time-series plot
ggplotly(ShoppingAreaShopliftingMonthly)I have decided to narrow down the scope of my analysis so that I can examine the data more thoroughly and extract more significant information. To achieve this, I divided the ShoppingArea data into separate areas by identifying unique street IDs to distinguish each shopping area. As a result, I was able to identify five distinct shopping areas for analysis.
# Creating a dataframe to display each unique Shopping Area Street ID
UniqueShoppingAreas <- as.data.frame(unique(ShoppingArea$street_id))
UniqueShoppingAreas## unique(ShoppingArea$street_id)
## 1 2153173
## 2 2153025
## 3 2153123
## 4 2152922
## 5 2153395Here, I am processing the data into separate dataframes for each of the shopping.
# Filtering data for the first shopping area
ShoppingArea1 <- ShoppingArea %>% filter(street_id == c('2153173'))
ShoppingArea1 <- ShoppingArea1 %>%
mutate(street_id = "ShoppingArea1") # Renaming the street_id to make analysis clear
ShoppingArea1Count <- ShoppingArea1 %>%
group_by(category, street_id) %>% # Grouping by category and street_id
summarise(crime_count = n()) # Finding Crime frequency ## `summarise()` has grouped output by 'category'. You can override using the
## `.groups` argument.# Filtering data for the second shopping area
ShoppingArea2 <- ShoppingArea %>% filter(street_id == c('2153025'))
ShoppingArea2 <- ShoppingArea2 %>%
mutate(street_id = "ShoppingArea2") # Renaming the street_id
ShoppingArea2Count <- ShoppingArea2 %>%
group_by(category, street_id) %>%
summarise(crime_count = n()) # Calculating Crime frequency ## `summarise()` has grouped output by 'category'. You can override using the
## `.groups` argument.# Filtering data for the third shopping area
ShoppingArea3 <- ShoppingArea %>% filter(street_id == c('2153395'))
ShoppingArea3 <- ShoppingArea3 %>%
mutate(street_id = "ShoppingArea3") # Renaming the street_id
ShoppingArea3Count <- ShoppingArea3 %>%
group_by(category, street_id) %>%
summarise(crime_count = n()) # Calculating Crime frequency ## `summarise()` has grouped output by 'category'. You can override using the
## `.groups` argument.# Filtering data for the fourth shopping area
ShoppingArea4 <- ShoppingArea %>% filter(street_id == c('2153123'))
ShoppingArea4 <- ShoppingArea4 %>%
mutate(street_id = "ShoppingArea4") # Renaming the street_id
ShoppingArea4Count <- ShoppingArea4 %>%
group_by(category, street_id) %>%
summarise(crime_count = n()) # Calculating Crime frequency ## `summarise()` has grouped output by 'category'. You can override using the
## `.groups` argument.# Filtering data for the fifth shopping area
ShoppingArea5 <- ShoppingArea %>% filter(street_id == c('2152922'))
ShoppingArea5 <- ShoppingArea5 %>%
mutate(street_id = "ShoppingArea5") # Renaming the street_id
ShoppingArea5Count <- ShoppingArea5 %>%
group_by(category, street_id) %>%
summarise(crime_count = n()) # Calculating Crime frequency ## `summarise()` has grouped output by 'category'. You can override using the
## `.groups` argument.# Combining All shopping area data frames including their new street IDs
AllShoppingAreas <- rbind(ShoppingArea1Count,ShoppingArea2Count,ShoppingArea3Count,ShoppingArea4Count,ShoppingArea5Count)I have analysed the data and created a stacked bar chart to show the distribution of crimes in five different shopping areas. The chart indicates that ShoppingArea4 had the highest number of crimes, with 105 shoplifting incidents and 43 violent crimes occurring in 2023. This is a significant difference compared to the other shopping areas. ShoppingArea1 had the second-highest number of crimes, with only 64 shoplifting incidents and 13 violent crimes. The difference between ShoppingArea4 and all other shopping areas is very clear. ShopliftingArea3 looks to have a very small number of crime compared to the rest of the shopping areas. Plotly documentation used for code can be accessed by referring to [5].
# Using Plotly to generate an interactive stacked bar chart
plot_ly(AllShoppingAreas, x = ~street_id, y = ~crime_count, color = ~category, type = "bar", colors = colour_palette) %>%
layout(title = "Distribution of Crime Categories Across Shopping Areas",
xaxis = list(title = "Shopping Area"),
yaxis = list(title = "Crime Count"),
barmode = "stack")However, the bar chart does not allow us to see the graphical location of each shopping area. Therefore, I will introduce a map displaying the graphical location of each area.
The map displays two shopping areas, ShoppingArea4 and ShoppingArea2, situated in the center of Colchester. This is not unexpected since the map displayed in the previous section demonstrated that the centre of Colchester has the highest crime rate. ShoppingArea3 can be observed to be situated outside of the city center within a small shopping area that comprises Halfords and Pets at Home. The marker for this area is less dense, indicating a lower number of crimes. On the other hand, ShoppingArea1 is located in a much larger retail park with numerous shops and restaurants, as seen on the map, and has a more dense marker. This suggests that larger shopping areas tend to experience more crime.
# Merging data
AllShoppingAreas <- rbind(ShoppingArea1,ShoppingArea2,ShoppingArea3,ShoppingArea4,ShoppingArea5)
# Grouping data
map <- AllShoppingAreas %>%
group_by(street_name, category)
# Plotting Map for each shopping area
m <- leaflet(map) %>%
addTiles() %>% setView(0.901681, 51.89524, zoom = 14) %>%
addCircleMarkers(popup = ~street_id, radius = 7, color = '#F53646')## Assuming "long" and "lat" are longitude and latitude, respectivelymI have created a plot that compares the average temperature in Colchester with that of ShoppingArea4. I chose to use ShoppingArea4 because it had the highest frequency of crime occurrences among all the shopping areas I analysed. I have also introduced loess smoothing in this plot making it easier to understand any trends in the data.
In order to merge the data from temp2023 with the ShoppingArea4 dataset, some data manipulation was required. Initially, the date attribute in temp2023 was in a daily format, while the date attribute in ShoppingArea4 was in a Year-Month (Y-M) format. To make them compatible, I reformatted the date attribute in temp2023 to match the Y-M format. However, this was insufficient to merge the data accurately since the TemperatureCAvg attribute in temp2023 was still based on the daily format. To resolve this issue, I grouped the entire temp2023 dataset by the new monthly data format and calculated the mean average temperature for each month. This ensured that the TemperatureCAvg attribute was in the same format as the crime_count attribute from the ShoppingArea4 dataset, making it possible to merge the data accurately.
The red line in the graph represents the average temperature. The data points that are closer to this line show a consistent trend. On the other hand, the blue data points representing the crime count in ShoppingArea4 are further away from the line, indicating an inconsistent trend. If we compare both lines, we can see a clear correlation between temperature and crime counts after September 2023, as both values show a similar dip. However, the trends at the beginning of the year are not as similar, with a gradual increase in crimes and a faster increase in temperature.
# Reformatting the 'Date' attribute for the temp2023 dataset to monthly
AvgTempPerMonth <- temp2023 %>%
group_by(date = format(as.Date(temp2023$Date, '%Y-%m-%d'), "%Y-%m")) %>%
summarise(avg_temp = mean(TemperatureCAvg)) # Calculating the average temperature for each month
# Grouping the ShoppingArea4 data by month
ShoppingAreaMonthly <- ShoppingArea4 %>%
group_by(date) %>%
summarise(crime_count = n())
# Merging data frames by date
TempShoppingArea4 <- merge(AvgTempPerMonth, ShoppingAreaMonthly, by = "date")
# Creating the time-series plot to compare temperature with crime trends
CrimeAvgTempPerMonth <- ggplot(TempShoppingArea4, aes(x = date)) +
geom_smooth(aes(y = avg_temp, colour = 'Average Temperature', group = 1), method = "loess") + # Using smoothing
geom_point(aes(y = avg_temp, colour = 'Average Temperature')) + # Plotting data points
geom_smooth(aes(y = crime_count, colour = 'Crime Count', group = 1), method = "loess") + # Using smoothing
geom_point(aes(y = crime_count, colour = 'Crime Count')) + # Plotting data points
labs(title = "ShoppingArea4 With Average Temperature Per Month", # Adding titles
x = "Month",
y = "Value") +
theme(axis.text.x = element_text(angle = 45, hjust = 1)) # Adjusting the angle for the dates to enhance readibility
CrimeAvgTempPerMonth## `geom_smooth()` using formula = 'y ~ x'
## `geom_smooth()` using formula = 'y ~ x'
I have generated another plot to compare with the previous one. This plot displays the average temperature along with all shopping area crimes. It is evident that there is a gradual increase in both temperature and shoplifting cases during the first six months, and they both follow a similar trend. However, after August, both the temperature and crimes decrease, but the temperature decrease is more gradual than the sudden dip in crimes. Overall, the ShoppingArea4 plot had similar trends between temperature and crime at the end of the year with a sudden fall in both cases. Whereas, the plot with all shopping areas had similar trends at the beginning of the year.
# Counting crime occurrences for all shopping areas together
ShoppingAreaMonthly <- ShoppingArea %>%
group_by(date) %>%
summarise(crime_count = n())
# Merging temperature data with all shopping area data
TempAllShoppingAreas <- merge(AvgTempPerMonth, ShoppingAreaMonthly, by = "date")
# Creating the time-series plot to compare all shopping area crime occurrences with the average temperature
CrimeAvgTempPerMonth <- ggplot(TempAllShoppingAreas, aes(x = date)) +
geom_smooth(aes(y = avg_temp, colour = 'Average Temperature', group = 1), method = "loess") + # Adding smoothing
geom_point(aes(y = avg_temp, colour = 'Average Temperature')) + # Adding data points for average temperature
geom_smooth(aes(y = crime_count, colour = 'Crime Count', group = 1), method = "loess") + # Adding smoothing
geom_point(aes(y = crime_count, colour = 'Crime Count')) + # Adding data points for crime counts
labs(title = "All Shopping Areas With Average Temperature Per Month", # Adding titles
x = "Month",
y = "Value") +
theme_minimal() +
theme(axis.text.x = element_text(angle = 45, hjust = 1)) # Adjust the date angles for better readability
# Displaying graph
CrimeAvgTempPerMonth## `geom_smooth()` using formula = 'y ~ x'
## `geom_smooth()` using formula = 'y ~ x'
Shopping Area Shoplifting Crime Outcome Analysis
In this section, my objective is to analyse the “crime_status” attribute of the “crime23” dataset. Upon inspecting the dataset, I noticed that the “outcome_status” attribute had 677 missing values. To address this issue, I decided to replace these missing values with “No Outcome Status”. This allowed me to visualize the number of incomplete rows in the “outcome_status” attribute.
# Counting NAs for outcome status
OutcomeStatusNA <- sum(is.na(crime23$outcome_status))
OutcomeStatusNA## [1] 677# Replacing NAs with 'No Outcome Status'
crime23 <- replace_na(crime23, list(outcome_status = 'No Outcome Status'))The dot plot displayed below illustrates the distribution of outcomes for all crimes. As per the graph, the highest percentage of 53.1% shows that investigations were concluded without identifying any suspects. In 17.7% of the cases, investigators were unable to prosecute the suspect. Meanwhile, 9.38% of crime outcomes were left incomplete in the dataset.
# Counting each outcome_status frequency for shopping areas
CrimeOutcome <- ShoppingArea %>%
group_by(outcome_status) %>%
summarize(count = n())
# Calculating the sum to calculate the percentage distribution for the dot plot
TotalCount <- sum(CrimeOutcome$count)
# Calculating the percentage distribution
OutcomeDistribution <- CrimeOutcome %>% mutate(Percentage = round((count / TotalCount) * 100, digits = 2))
# Creating the dot plot
Dotplot <- ggplot(OutcomeDistribution, aes(x = outcome_status, y = Percentage)) +
geom_point(shape = 4, size = 3, colour = '#F53646') +
labs(title = "Percentage Distribution of Crime Outcomes", x = "Outcome Status", y = "Percentage") +
coord_flip()
# Using ggplotly to create an interactive dot plot
ggplotly(Dotplot)The scatter plot below illustrates the relationship between Shoplifting Counts in Shopping Areas and Complete Investigations with no suspects per month. Although there is no linear relationship between these two variables, there is a stronger relationship between the outcome status and shoplifting counts closer to the middle of the plot. The plot shows that as the number of shoplifting offences increases, the number of investigations also increases. Suggesting that there might be a connection between the frequency of shoplifting and complete investigations with no suspect identified.
# Calculating the number of shoplifting offences for all shopping areas
ShopliftingCrimeCount <- ShoppingArea %>%
filter(category == 'shoplifting') %>%
group_by(date) %>%
summarize(count = n())
# Calculating the number of 'Investigation complete; no suspect identified' outcomes by date
CompleteInvestigationCount <- ShoppingArea %>%
filter(outcome_status == 'Investigation complete; no suspect identified') %>%
group_by(date) %>%
summarize(count = n())
# Creating the scatter plot
ShopliftingCrimeScatter <- ggplot() +
geom_point(data = ShopliftingCrimeCount, aes(x = count, y = CompleteInvestigationCount$count), size = 5, shape = 1, colour = '#F53646') +
labs(x = "Number of Shoplifting Offences", y = "Number of complete investigations", title = 'Shoplifting Counts Against Their
Complete Investigations with No Suspect Identified Per Month')
# Using ggplotly to create an interactive scatter plot
ggplotly(ShopliftingCrimeScatter)Here, I have introduced a density plot for the number of complete investigations per shopping area. However, ShoppingArea3 only includes 1 observation after the data filtering, therefore, a density graph is not possible for this location and will be dropped from this analysis. ShoppingArea5 is right skewed therefore, the mean is greater than the median. ShoppingArea1 has a wider frequency of outcomes where no suspects were identified within the range of 0-8 the density curve is pronounced here then dips at 12. The other shopping areas appear to be more spread-out on occurrences where no suspect was identified.
# Filtering each shopping area by the 'Investigation complete; no suspect identified' outcome_status
# and also counting the number of occurrences for this outcome.
ShoppingArea1CompleteInvestigations <- ShoppingArea1 %>% filter(outcome_status == 'Investigation complete; no suspect identified') %>%
group_by(date, street_id) %>%
summarise(count = n());## `summarise()` has grouped output by 'date'. You can override using the
## `.groups` argument.ShoppingArea2CompleteInvestigations <- ShoppingArea2 %>% filter(outcome_status == 'Investigation complete; no suspect identified') %>%
group_by(date, street_id) %>%
summarise(count = n());## `summarise()` has grouped output by 'date'. You can override using the
## `.groups` argument.ShoppingArea4CompleteInvestigations <- ShoppingArea4 %>% filter(outcome_status == 'Investigation complete; no suspect identified') %>%
group_by(date, street_id) %>%
summarise(count = n());## `summarise()` has grouped output by 'date'. You can override using the
## `.groups` argument.ShoppingArea5CompleteInvestigations <- ShoppingArea5 %>% filter(outcome_status == 'Investigation complete; no suspect identified') %>%
group_by(date,street_id) %>%
summarise(count = n());## `summarise()` has grouped output by 'date'. You can override using the
## `.groups` argument.# Binding each data frame together for the density plot
ShoppingAreaCountsAll <- rbind(ShoppingArea1CompleteInvestigations, ShoppingArea2CompleteInvestigations, ShoppingArea4CompleteInvestigations, ShoppingArea5CompleteInvestigations)
# Creating the density plot
DensityOutcomeShoppingArea <- ggplot(ShoppingAreaCountsAll, aes(x = count, fill = street_id)) +
geom_density(alpha = 0.6) + # Setting alpha value for a clear density plot
labs(title = "Density Plot of Complete Investigations With No Suspect Identified
By Shopping Area",
x = "No Suspect Counts",
y = "Density",
fill = "Season") +
scale_fill_manual(values = c('#F53646', '#4F758B', '#AAF1E6', '#821B04', '#FFC78C')) # Adding the colour palette
DensityOutcomeShoppingArea
The following histogram displays the number of investigations that were completed in 2023, but where no suspect was identified. The x-axis represents the months of the year, and the y-axis represents the number of such investigations. Upon observing the graph, we can see that the months of May and August had the highest number of investigations with no identified suspects. However, the graph shows fluctuations throughout the year, with no clear trends.
# Creating histogram plot
ShoppingArea_per_month <- ggplot(ShoppingArea4CompleteInvestigations, aes(x = date, y = count)) +
geom_bar(stat = "identity", color = 'black', fill = "#F53646") +
labs(title = "ShoppingArea4 Crimes Where No Suspect was Identified Per Month",
x = "Month",
y = "Number of Offences")
# Displaying plot
ShoppingArea_per_month
The Violin plot below shows the number of crimes committed in different shopping areas where no suspect was identified per month. The y-axis represents the number of crimes, while the x-axis displays the shopping areas. Shopping areas 1 and 4 have narrower violins, which indicate lower density, while shopping areas 2 and 5 have wider violins. Shopping areas 1 and 4 have a wider range of cases throughout the year, while shopping areas 2 and 5 have a shorter range, indicating fewer cases.
# Creating violin plot
ggplot(ShoppingAreaCountsAll, aes(x = factor(street_id), y = count)) +
geom_violin(fill = '#F53646' , alpha = 0.6) +
labs(title = "All ShoppingArea Crimes Where No Suspect was Identified Per Month",
y = "Count",
x = "Shopping Area") +
stat_summary(fun.y = median, geom='point') # Adding stat summary so median is visible ## Warning: The `fun.y` argument of `stat_summary()` is deprecated as of ggplot2 3.3.0.
## ℹ Please use the `fun` argument instead.
## This warning is displayed once every 8 hours.
## Call `lifecycle::last_lifecycle_warnings()` to see where this warning was
## generated.
Shopping Area Correlation analysis
I conducted a correlation analysis using all shopping areas to examine the relationship between weather conditions and shoplifting incidents. The resulting Correlation Matrix displays the correlations between shoplifting offences and weather conditions. I specifically focused on shoplifting incidents in April, May, June, and July, as from the earlier time series plot I found a gradual decrease in crime frequency during this time of the year. The main aim of this correlation analysis was to determine whether temperature had an impact on the frequency of shoplifting offences during this particular time of year.
Each square in the matrix illustrates the correlation between shoplifting and various weather conditions. The coefficient value closer to 1 indicates an increase in shoplifting incidents with that particular weather condition, while a coefficient closer to -1 suggests a decrease in shoplifting incidents with that particular weather condition. A coefficient value of 0 indicates no correlation between shoplifting and the weather.
In this correlation matrix, darker red boxes suggest a strong positive correlation, while darker blue boxes indicate a strong negative correlation.
According to the correlation matrix, there is a strong negative correlation between shoplifting and temperature, which implies that shoplifting cases decrease as temperature increases. On the other hand, there is a positive correlation between shoplifting and average humidity, wind gust, and wind speed, suggesting that shoplifting increases with these weather attributes. Other weather factors such as precipitation and pressure have no significant relationship with shoplifting frequency. Overall, the correlation matrix suggests that colder temperatures, higher humidity, and wind speed are associated with more shoplifting cases during the analysed time frame.
# Filtering shopping areas for shoplifting offences
Shoplifting <- ShoppingArea %>% filter(category == 'shoplifting')
# Filtering by date and counting the number of shoplifting offences
Shoplifting <- Shoplifting %>% filter(date == c('2023-04', '2023-05', '2023-06', '2023-07')) %>%
group_by(date) %>%
summarise(ShopliftingCount = n())## Warning: There was 1 warning in `filter()`.
## ℹ In argument: `date == c("2023-04", "2023-05", "2023-06", "2023-07")`.
## Caused by warning in `date == c("2023-04", "2023-05", "2023-06", "2023-07")`:
## ! longer object length is not a multiple of shorter object length# Reformatting the weather date attribute to year-month
temp_new <- temp2023
temp_new$Date <- format(as.Date(temp2023$Date, '%Y-%m-%d'), "%Y-%m")
# Filtering by date and calculating the average for the weather data
temp_new_monthly <- temp_new %>% filter(Date == c('2023-04', '2023-05', '2023-06', '2023-07')) %>%
group_by(Date) %>%
summarise(across(where(is.numeric), mean))## Warning: There was 1 warning in `filter()`.
## ℹ In argument: `Date == c("2023-04", "2023-05", "2023-06", "2023-07")`.
## Caused by warning in `Date == c("2023-04", "2023-05", "2023-06", "2023-07")`:
## ! longer object length is not a multiple of shorter object length# Merging the filtered weather data with the filtered shoplifting data frame
merged_ShoppingArea <- merge(Shoplifting, temp_new_monthly, by.x = "date", by.y = "Date")
# Creating a vector containing the names of all numeric attributes for the correlation matrix
numerical_cols <- c('ShopliftingCount', 'TemperatureCAvg', 'TemperatureCMax', 'TemperatureCMin',
'TdAvgC', 'HrAvg', 'WindkmhInt', 'WindkmhGust', 'PresslevHp',
'Precmm', 'TotClOct', 'SunD1h', 'VisKm')
# Computing the correlation between variables
correlation_matrix <- cor(merged_ShoppingArea[, numerical_cols])
# Creating the correlation matrix for All shopping areas
ggcorrplot(correlation_matrix, type = "lower",
title = "Correlation Matrix of Shoplifting Frequency
in Shopping Areas With Weather Data",
colors = c("#5184C8", "white", "#F53646")) +
theme(plot.title = element_text(size=10)) # Adjusting title size## Warning: `aes_string()` was deprecated in ggplot2 3.0.0.
## ℹ Please use tidy evaluation idioms with `aes()`.
## ℹ See also `vignette("ggplot2-in-packages")` for more information.
## ℹ The deprecated feature was likely used in the ggcorrplot package.
## Please report the issue at <https://github.com/kassambara/ggcorrplot/issues>.
## This warning is displayed once every 8 hours.
## Call `lifecycle::last_lifecycle_warnings()` to see where this warning was
## generated.
I have generated another correlation matrix that compares the weather to shoplifting cases, but this time only for the ShoppingArea4 location. In this correltaion matrix, the orange squares are positive correlations while blue squares represent negative correlations. The correlation matrix indicates some weak positive relationships within temperature attributes. This implies that shoplifting tends to increase as the temperature increases. This finding is interesting as it contradicts the previous correlation matrix that considered all the shopping centres, which showed that the shoplifting cases decreased as the temperature increased. Therefore, it would be worth analysing the correlation of another shopping centre located outside the centre of Colchester to see if the results differ.
# Filtering shopping areas for ShoppingArea4 offences
Shoplifting <- ShoppingArea4 %>% filter(category == 'shoplifting')
# Filtering by date and counting the number of shoplifting offences
Shoplifting <- Shoplifting %>% filter(date == c('2023-04', '2023-05', '2023-06', '2023-07')) %>%
group_by(date) %>%
summarise(ShopliftingCount = n())## Warning: There was 1 warning in `filter()`.
## ℹ In argument: `date == c("2023-04", "2023-05", "2023-06", "2023-07")`.
## Caused by warning in `date == c("2023-04", "2023-05", "2023-06", "2023-07")`:
## ! longer object length is not a multiple of shorter object length# Merging the filtered weather data with the filtered shoplifting data frame
merged_ShoppingArea4 <- merge(Shoplifting, temp_new_monthly, by.x = "date", by.y = "Date")
# Computing the correlation between variables
correlation_matrix <- cor(merged_ShoppingArea4[, numerical_cols])
# Creating the correlation matrix for ShoppingArea4
ggcorrplot(correlation_matrix, type = "lower",
title = "Correlation Matrix of Shoplifting Frequency
in ShoppingArea4 With Weather Data",
colors = c("#5184C8", "white", "#FFC78C")) +
theme(plot.title = element_text(size=10))
This is the correlation matrix for a shopping area situated outside the centre of Colchester (ShoppingArea1). In this correltaion matrix, the purple squares are postive correlations while blue squares represent negative correlations. It is evident from the matrix that this shopping area is similar to the first correlation matrix, which contains data on all shopping areas. The matrix also shows that the shopping counts have a negative correlation with the temperature attributes. Indicating that as temperature increases, shoplifting decreases for ShoppingArea1.
# Filtering shopping areas for ShoppingArea1 offences
Shoplifting <- ShoppingArea1 %>% filter(category == 'shoplifting')
# Filtering by date and counting the number of shoplifting offences
Shoplifting <- Shoplifting %>% filter(date == c('2023-04', '2023-05', '2023-06', '2023-07')) %>%
group_by(date) %>%
summarise(ShopliftingCount = n())
# Merging the filtered weather data with the filtered shoplifting data frame
merged_ShoppingArea1 <- merge(Shoplifting, temp_new_monthly, by.x = "date", by.y = "Date")
# Computing the correlation between variables
correlation_matrix <- cor(merged_ShoppingArea1[, numerical_cols])
# Creating the correlation matrix for ShoppingArea1
ggcorrplot(correlation_matrix, type = "lower",
title = "Correlation Matrix of Shoplifting Frequency
in ShoppingArea1 With Weather Data",
colors = c("#5184C8", "white", "#AB274F")) +
theme(plot.title = element_text(size=10))
Conclusion
The analysis of the policing dataset using various data visualization techniques has provided valuable insights into crime patterns in Colchester, with a specific focus on shopping areas. The findings reveal that shopping areas have the highest crime rates, with shoplifting being the most prevalent type of crime in these areas. It has been determined that Shopping Area 4, situated in the centre of Colchester, has the highest crime rate among all shopping centers. Upon examining a map, I noticed a pattern that smaller shopping areas tend to have lower crime rates compared to larger ones.
Furthermore, the average temperature also follows the general shoplifting trends throughout 2023. It was found that 53.1% of crimes within shopping areas resulted in no suspect being identified, which was at its highest in August. The correlation analysis also identified how all shopping and singled-out shopping areas correlate with the weather conditions. The results revealed that ShoppingArea4 shows that as the temperature increases, so does shoplifting, while for ShoppingArea1, shoplifting decreases while the temperature increases. The insights obtained from this analysis aim to help improve policing in Colchester and reduce crime in the shopping areas of Colchester.
References:
Association LG. Crime and disorder in Colchester, 2023 Q3 (12 months ending) [Internet]. lginform.local.gov.uk. [cited 2024 Apr 24]. Available from: https://lginform.local.gov.uk/reports/view/lga-research/lga-research-report-police-recorded-crime?mod-area=E07000071&mod-group=AllDistrictInRegion_East&mod-type=namedComparisonGroup
About | data.police.uk [Internet]. Police.uk. 2016. Available from: https://data.police.uk/about/#location-anonymisation
Scrapping meteorological (Synop) data from the Ogimet webpage — meteo_ogimet [Internet]. bczernecki.github.io. Available from: https://bczernecki.github.io/climate/reference/meteo_ogimet.html
plotlygraphs. Pie Charts [Internet]. Plotly.com. plotlygraphs; 2015. Available from: https://plotly.com/r/pie-charts/
Bar [Internet]. plotly.com. Available from: https://plotly.com/r/bar-charts/